Currently, methods for Raman gain control in a hybrid amplifier are mostly control methods based on feed-forward plus feedback. Such control methods have been proved to be effective in practical engineering applications, but they are not perfect in accuracy of Raman gain control. Therefore, some techniques have been proposed to improve accuracy of Raman gain control by using gain calibration methods, which may improve the accuracy of Raman gain control for a single hybrid amplifier in the laboratory. However, when plural hybrid amplifiers are connected in cascade in practical engineering applications, the gain calibration methods may fail. Referring to FIG. 1, three hybrid amplifiers, which are labeled as Hybrid 1, Hybrid 2, and Hybrid 3 respectively, are connected in cascade, each including a Raman fiber amplifier (RFA) and an Erbium-Doped Fiber Amplifier (EDFA). The gain calibration methods assume the input light for a hybrid amplifier remains constant, and the gain calibration is realized based on change of the input light. When the hybrid amplifiers are connected in cascade, the Raman fiber amplifier and the EDFA are controlled synchronously with control timing shown in FIG. 2, which illustrates a timing diagram for pump-starting of the first and second hybrid amplifiers in the cascade. For the first hybrid amplifier, control of the hybrid amplifier includes determining whether the input light is stable according to change of slope of the input light power. If it is determined that the input light power is stable, the Raman and the EDFA may start pumping simultaneously. Then, the Raman enters into an automatic gain calibration procedure A, after which the Raman may switch to a target mode, i.e. an AGC (automatic gain control) mode. For the EDFA, it may enter into the AGC mode quickly after it starts pumping. However, due to change of the output light power of the Raman after the automatic gain calibration is completed, the EDFA generates also a change in its output light power. The output of the EDFA is the output of the first hybrid amplifier. Thus, a problem arises when a plurality of amplifiers are used in cascade. Specifically, the output light power of a hybrid amplifier is used for gain calibration of a next hybrid amplifier, while a precondition for correct gain calibration is that the input light entering into the Raman must remain stable. In FIG. 2, the input light of the first hybrid amplifier start changing from scratch at a timing 1, and the changing is relatively slow. At the timing 1, the hybrid amplifier checks whether the input light becomes stable. When it is determined that the input light has been stable at a timing 2, the Raman and the EDFA in the first hybrid amplifier start pumping simultaneously. For the second hybrid amplifier, as the output light of the EDFA in the first hybrid amplifier is unstable from the pump-starting timing 2 to a timing 3, the Raman in the second hybrid amplifier determines whether the input light is stable during the timing 2 to the timing 3, and then implements the gain calibration during the timing 3 to a timing 4. As the output light of the EDFA in the first hybrid amplifier changes when the Raman in the second hybrid amplifier is implementing the automatic gain calibration B (from the timing 3 to the timing 4), the automatic gain calibration of the Raman in the second hybrid amplifier cannot result in a correct calibration, causing an obvious deviation in control accuracy of the Raman. Also, it may be seen from FIG. 2 that due to the cascade of the plurality of hybrid amplifiers, a further problem may arise that a next hybrid amplifier starts pumping later than a last hybrid amplifier does.